[RFC] [Question] memset.S old code stuff

[RFC] [Question] memset.S old code stuff

[Bradley Morgan]

Hey folks, sorry for the crazy asm emails, I see that in memset.S in
arch/arm64/lib/memset.S, the code is incredibly old code, like 2013 old
(heh).


I was wondering if I could say, adapt it from optimized-routines (same
treatment as memcpy.S essentially.
Thanks!

Re: [RFC] [Question] memset.S old code stuff

[Catalin Marinas]

On Thu, Jul 23, 2026 at 12:17:15AM +0100, Bradley Morgan wrote:
> Hey folks, sorry for the crazy asm emails, I see that in memset.S in
> arch/arm64/lib/memset.S, the code is incredibly old code, like 2013 old
> (heh).
> 
> I was wondering if I could say, adapt it from optimized-routines (same
> treatment as memcpy.S essentially.

I think Robin attempted until he realised the use of the Q registers in
the Cortex Strings library, mentioned here:

https://lore.kernel.org/all/cover.1571073960.git.robin.murphy@arm.com/

So that's where Cortex Strings diverged from what we can use in the
kernel.

-- 
Catalin

Re: [RFC] [Question] memset.S old code stuff

[Mark Rutland]

On Tue, Aug 04, 2026 at 04:57:54PM +0100, Catalin Marinas wrote:
> On Thu, Jul 23, 2026 at 12:17:15AM +0100, Bradley Morgan wrote:
> > Hey folks, sorry for the crazy asm emails, I see that in memset.S in
> > arch/arm64/lib/memset.S, the code is incredibly old code, like 2013 old
> > (heh).
> > 
> > I was wondering if I could say, adapt it from optimized-routines (same
> > treatment as memcpy.S essentially.
> 
> I think Robin attempted until he realised the use of the Q registers in
> the Cortex Strings library, mentioned here:
> 
> https://lore.kernel.org/all/cover.1571073960.git.robin.murphy@arm.com/

We also subsequently looked at moving away from Cortex-strings entirely,
e.g.

  https://lore.kernel.org/linux-arm-kernel/20230314115030.347976-1-mark.rutland@arm.com/

... but there were open questions on the semantics, and no clear path
forwards at the time.

Writing the code isn't the hard part. We need to get wider agreement on
the semantics, test on a variety of platforms, etc.

For now, I think it'd be best to leave as-is, since it's a time sink for
reviewers.

Mark.

Re: [RFC] [Question] memset.S old code stuff

[Bradley Morgan]

On 4 August 2026 18:20:43 BST, Mark Rutland <mark.rutland@arm.com> wrote:
>On Tue, Aug 04, 2026 at 04:57:54PM +0100, Catalin Marinas wrote:
>> On Thu, Jul 23, 2026 at 12:17:15AM +0100, Bradley Morgan wrote:
>> > Hey folks, sorry for the crazy asm emails, I see that in memset.S in
>> > arch/arm64/lib/memset.S, the code is incredibly old code, like 2013
>old
>> > (heh).
>> > 
>> > I was wondering if I could say, adapt it from optimized-routines (same
>> > treatment as memcpy.S essentially.
>> 
>> I think Robin attempted until he realised the use of the Q registers in
>> the Cortex Strings library, mentioned here:
>>

I would say scalar port??

>> https://lore.kernel.org/all/cover.1571073960.git.robin.murphy@arm.com/
>
>We also subsequently looked at moving away from Cortex-strings entirely,
>e.g.
>
>  https://lore.kernel.org/linux-arm-kernel/20230314115030.347976-1-mark.rutland@arm.com/
>
>... but there were open questions on the semantics, and no clear path
>forwards at the time.
>
>Writing the code isn't the hard part. We need to get wider agreement on
>the semantics, test on a variety of platforms, etc.
>
>For now, I think it'd be best to leave as-is, since it's a time sink for
>reviewers.

Ack. Again, a scalar port is iffy, not 100%. 

That would require time.


>Mark.
>

Thanks!